🐛 Describe the bug

Running this code throws an exception in bazel build

#include <torch/torch.h>
using namespace torch::nn;

int main() {
    Linear x(128, 64);
    x->to(torch::kInt32);
    return 0;
}

C++ exception with description "data set to a tensor that requires gradients must be floating point or complex dtype

This is tested in CMake build by this test:

Versions

master 7d17e3b

cc @jbschlosser